import numpy as np
import pandas as pd
import math
import matplotlib.pyplot as plt
import seaborn as sns
from scipy.stats import linregress, zscore

# Load and convert types
# (a) Import
# (b) Parse

car_sales_df = pd.read_csv("car_sales_data.csv")
display(car_sales_df.head())
display(car_sales_df.dtypes)

year              int64
make             object
model            object
trim             object
body             object
transmission     object
vin              object
state            object
condition       float64
odometer        float64
color            object
interior         object
seller           object
mmr             float64
sellingprice    float64
saledate         object
dtype: object

# Data Testing/Cleaning
display(car_sales_df.isna().sum())

display(car_sales_df[car_sales_df.isna().any(axis=1)].head())

car_sales_df = car_sales_df.dropna()

# Convert sale date to datetime object

car_sales_df["saledate"] = pd.to_datetime(
    car_sales_df["saledate"].str.split(" \(").str[0],  # apply split elementwise
    format="%a %b %d %Y %H:%M:%S GMT%z",
    utc=True
)

# Cars are released in the previous year
car_sales_df["year"] = pd.to_datetime(car_sales_df["year"] - 1, format="%Y", utc=True)

display(car_sales_df.head())

year                0
make            10301
model           10399
trim            10651
body            13195
transmission    65352
vin                 4
state               0
condition       11820
odometer           94
color             749
interior          749
seller              0
mmr                38
sellingprice       12
saledate           12
dtype: int64

# Describe the dataframe
car_sales_df.describe()

# Number of different manufacturers and their representation in the dataset
print(f"Number of Manufacturers:")
display(len(car_sales_df["make"].unique()))

print("\nCount of Models by Manufacturers:")
display(car_sales_df.groupby('make')["model"].unique())

print("\nNumber of Unique Car Models:")
display(len(car_sales_df["model"].unique()))

print("\nMost Popular Model")
display(car_sales_df.groupby("model")["make"].count().idxmax(), car_sales_df.groupby("model")["make"].count().max())

Number of Manufacturers:

53

Count of Models by Manufacturers:

make
Acura            [ILX, MDX, TSX, TL, RDX, ZDX, RSX, RL, CL, Int...
Aston Martin                             [V8 Vantage, DB9, Rapide]
Audi             [A4, A6, Q5, A3, SQ5, S5, A8, TTS, S4, A7, A5,...
BMW              [3 Series, 6 Series Gran Coupe, M5, 6 Series, ...
Bentley          [Continental GT Speed, Continental Flying Spur...
Buick            [Enclave, LaCrosse, Rendezvous, LeSabre, Centu...
Cadillac         [ELR, SRX, Escalade, CTS Coupe, Escalade ESV, ...
Chevrolet        [Cruze, Camaro, Impala, Suburban, Malibu, Silv...
Chrysler         [200, 300, Town and Country, Sebring, PT Cruis...
Daewoo                                                    [Nubira]
Dodge            [Avenger, Journey, Charger, Grand Caravan, Nit...
FIAT                                             [500, 500L, 500e]
Ferrari                        [California, F430, 360, 458 Italia]
Fisker                                                     [Karma]
Ford             [Fusion, Escape, Edge, Focus, F-350 Super Duty...
GMC              [Terrain, Yukon, Sierra 1500, Acadia, Yukon XL...
Geo                                                 [Prizm, Metro]
HUMMER                                   [H3, H2, H2 SUT, H3T, H1]
Honda            [Accord, CR-V, Civic, Fit, Pilot, Odyssey, Cro...
Hyundai          [Elantra, Santa Fe, Genesis, Equus, Sonata, So...
Infiniti         [G Coupe, G Sedan, FX, M, JX, G Convertible, Q...
Isuzu            [Rodeo, Ascender, Axiom, Rodeo Sport, Trooper,...
Jaguar           [XF, XJ, XK, XJ-Series, S-Type, X-Type, F-TYPE...
Jeep             [Wrangler, Compass, Grand Cherokee, Liberty, P...
Kia              [Sorento, Optima, K900, Rio, Soul, Forte, Spor...
Lamborghini                                             [Gallardo]
Land Rover       [LR4, Range Rover Evoque, Range Rover Sport, R...
Lexus            [RX 350, GS 350, ES 350, LS 460, IS 250, CT 20...
Lincoln          [MKX, MKZ, Navigator, Town Car, MKT, MKS, Mark...
Lotus                                                      [Exige]
MINI             [Cooper Clubman, Cooper, Cooper Countryman, Co...
Maserati         [GranTurismo, Quattroporte, GranTurismo Conver...
Mazda            [Mazda2, Mazda3, CX-9, Mazda5, Mazda6, CX-7, M...
Mercedes-Benz    [S-Class, C-Class, SLK-Class, E-Class, GLK-Cla...
Mercury          [Milan, Monterey, Mountaineer, Marauder, Grand...
Mitsubishi       [Outlander, Outlander Sport, Lancer, Galant, L...
Nissan           [Altima, Versa, Versa Note, 370Z, Juke, NV, Le...
Oldsmobile       [Alero, Intrigue, Cutlass Ciera, Silhouette, A...
Plymouth                                   [Neon, Prowler, Breeze]
Pontiac          [G5, G6, Grand Prix, Solstice, Grand Am, Sunfi...
Porsche          [Cayenne, 911, Boxster, Panamera, Cayman S, Ca...
Ram              [1500, 2500, 3500, C/V Cargo Van, Dakota, C/V ...
Rolls-Royce                                                [Ghost]
Saab                                        [9-3, 9-5, 9-7X, 9-2X]
Saturn           [VUE, Outlook, ION, Aura, L-Series, S-Series, ...
Scion                                   [FR-S, xB, tC, xD, iQ, xA]
Subaru           [Impreza WRX, Legacy, Forester, Impreza, Outba...
Suzuki           [XL-7, Grand Vitara, SX4, Kizashi, XL7, Forenz...
Tesla                                                    [Model S]
Toyota           [Corolla, Sienna, Yaris, Camry, Tacoma, FJ Cru...
Volkswagen       [Passat, CC, Jetta, Beetle, Jetta SportWagen, ...
Volvo            [S60, XC70, V60, XC60, S80, XC90, V50, S40, C3...
smart                                                     [fortwo]
Name: model, dtype: object

Number of Unique Car Models:

768

Most Popular Model

'Altima'

16346

# Histograms of numerical data
car_sales_df["year"].hist()
plt.title("Histogram of Year of Manufacture")
plt.show()

car_sales_df["odometer"].hist()
plt.title("Histogram of Mileage")
plt.show()

# The pretty plot
car_sales_df["sellingprice"].hist(bins=int(math.sqrt(car_sales_df["sellingprice"].size)))
plt.axvline(car_sales_df["sellingprice"].mean(), label=f"Average Car Price - {car_sales_df['sellingprice'].mean()}", color="black")
plt.axvline(car_sales_df["sellingprice"].quantile(.25), label=f"25th percentile - {car_sales_df['sellingprice'].quantile(.25)}", color="orange")
plt.axvline(car_sales_df["sellingprice"].quantile(.75), label=f"75th percentile - {car_sales_df['sellingprice'].quantile(.75)}", color="green")
plt.axvline(car_sales_df["sellingprice"].quantile(.99), label=f"99th percentile - {car_sales_df['sellingprice'].quantile(.99)}", color="red")
plt.title("Histogram of Car Prices")
plt.legend()
plt.show()

# Lets start by creating and visualizing our data
car_sales_df["age_at_sale"] = car_sales_df["saledate"] - car_sales_df["year"]

ages = car_sales_df["age_at_sale"].dt.days / 365.25

# Compute statistics
mean_val = ages.mean()
median_val = ages.median()
p25 = ages.quantile(0.25)
p75 = ages.quantile(0.75)

# Create histogram
fig, ax = plt.subplots(figsize=(10, 6))
ax.hist(ages, bins=50, color="skyblue", edgecolor="black", alpha=0.7)

# Add vertical lines for key stats
ax.axvline(mean_val, color="red", linestyle="--", linewidth=2, label=f"Mean: {mean_val:.1f}")
ax.axvline(median_val, color="green", linestyle="-.", linewidth=2, label=f"Median: {median_val:.1f}")
ax.axvline(p25, color="orange", linestyle=":", linewidth=2, label=f"25th pct: {p25:.1f}")
ax.axvline(p75, color="purple", linestyle=":", linewidth=2, label=f"75th pct: {p75:.1f}")

# Customize plot
ax.set_title("Distribution of Car Age at Sale (Years)", fontsize=14, fontweight="bold")
ax.set_xlabel("Age at Sale (Years)", fontsize=12)
ax.set_ylabel("Frequency", fontsize=12)
ax.grid(True, linestyle="--", alpha=0.5)
ax.legend()

# Display summary stats below the chart
summary_text = (
    f"Mean = {mean_val:.1f} Years\n"
    f"Median = {median_val:.1f} Years\n"
    f"25th percentile = {p25:.1f} Years\n"
    f"75th percentile = {p75:.1f} Years"
)
plt.figtext(0.75, 0.6, summary_text, fontsize=10, bbox=dict(facecolor='white', alpha=0.6))

plt.show()

# Create different dataframes based on the make of the cars
grouped_by_make = car_sales_df.groupby("make")

n = len(grouped_by_make)
cols = 3  # adjust as you like
rows = math.ceil(n / cols)

fig, axes = plt.subplots(rows, cols, figsize=(15, 4 * rows))
axes = axes.flatten()

results = []

# Create the plot for each make
for i, (make, group) in enumerate(grouped_by_make):
    group = group[group["age_at_sale"].dt.days != 0]
    if group.empty:
        continue

    # X axis will be the logrithm of how old the car is at sale (days). The Y axis is the selling price of the car
    x = group["age_at_sale"].dt.days.apply(math.log)
    y = group["sellingprice"]

    # Perform linear regression
    slope, intercept, r_value, p_value, std_err = linregress(x, y)
    results.append({
        "make": make,
        "slope": slope,
        "intercept": intercept,
        "r_squared": r_value**2,
        "p_value": p_value,
        "std_err": std_err,
        "samples": y.size
    })

    # Plot scatter + regression line
    ax = axes[i]
    ax.scatter(x, y, alpha=0.6)
    ax.plot(x, slope * x + intercept, color="red", lw=1.5)
    ax.set_title(f"{make}", fontsize=10, fontweight="bold")
    ax.set_xlabel("log(Age at Sale, Days)")
    ax.set_ylabel("Selling Price")
    ax.grid(True, linestyle="--", alpha=0.3)

# Hide unused subplots
for j in range(i + 1, len(axes)):
    axes[j].axis("off")

fig.suptitle("Age at Sale vs Selling Price by Make", fontsize=14, fontweight="bold")
fig.tight_layout(rect=[0, 0, 1, 0.97])
plt.show()

# Convert results to DataFrame and display
results_df = pd.DataFrame(results)
results_df = results_df.sort_values("p_value").reset_index(drop=True)
display(results_df.query("samples >= 30").sort_values(by="make"))

/Users/zacharygottlieb/anaconda3/lib/python3.11/site-packages/scipy/stats/_stats_mstats_common.py:182: RuntimeWarning: invalid value encountered in scalar divide
  slope = ssxym / ssxm
/Users/zacharygottlieb/anaconda3/lib/python3.11/site-packages/scipy/stats/_stats_mstats_common.py:196: RuntimeWarning: invalid value encountered in sqrt
  t = r * np.sqrt(df / ((1.0 - r + TINY)*(1.0 + r + TINY)))
/Users/zacharygottlieb/anaconda3/lib/python3.11/site-packages/scipy/stats/_stats_mstats_common.py:199: RuntimeWarning: invalid value encountered in scalar divide
  slope_stderr = np.sqrt((1 - r**2) * ssym / ssxm / df)

car_sales_df.dtypes

year            datetime64[ns, UTC]
make                         object
model                        object
trim                         object
body                         object
transmission                 object
vin                          object
state                        object
condition                   float64
odometer                    float64
color                        object
interior                     object
seller                       object
mmr                         float64
sellingprice                float64
saledate        datetime64[ns, UTC]
age_at_sale         timedelta64[ns]
dtype: object

# The columns that we want to check the outlier values for
numeric_columns = ("odometer", "sellingprice", "age_at_sale", "year")

# Create a new dataframe using only those columns. Transform the data to be correct (correct days be dividing by days in a year).
numeric_df = car_sales_df[["odometer", "sellingprice", "age_at_sale", "year"]].copy()
numeric_df["age_at_sale"] = numeric_df["age_at_sale"].dt.days / 365.25
numeric_df["year"] = numeric_df["year"].dt.year

display(numeric_df.describe())

n = len(numeric_columns)
cols = 2  # adjust as you like
rows = math.ceil(n / cols) * 2

fig, axes = plt.subplots(rows, cols, figsize=(15, 4 * rows))
axes = axes.flatten()

for i, col in enumerate(numeric_columns):
    curr = numeric_df[col]

    ax = axes[2 * i]
    ax.boxplot(curr, vert=False)
    ax.set_title(f"Boxplot of {col}", fontsize=10, fontweight="bold")
    ax.set_xlabel(col)
    ax.grid(True, linestyle="--", alpha=0.3)

    ax = axes[2 * i + 1]
    bins = int(math.sqrt(len(curr.unique())))
    ax.hist(curr, bins=bins, color="lightblue", edgecolor="black", alpha=0.7)

    # Add vertical lines for statistics
    ax.axvline(curr.mean(), color="black", linestyle="-", linewidth=1.2,
               label=f"Mean - {curr.mean():.2f}")
    ax.axvline(curr.quantile(0.25), color="orange", linestyle="--", linewidth=1.2,
               label=f"25th percentile - {curr.quantile(0.25):.2f}")
    ax.axvline(curr.quantile(0.75), color="green", linestyle="--", linewidth=1.2,
               label=f"75th percentile - {curr.quantile(0.75):.2f}")
    ax.axvline(curr.quantile(0.99), color="red", linestyle="--", linewidth=1.2,
               label=f"99th percentile - {curr.quantile(0.99):.2f}")

    # Titles and labels
    ax.set_title(f"Histogram of {col}", fontsize=10, fontweight="bold")
    ax.set_xlabel(col)
    ax.set_ylabel("Occurrences")
    ax.legend(fontsize=8)
    ax.grid(True, linestyle="--", alpha=0.3)

fig.suptitle("Age at Sale vs Selling Price by Make", fontsize=14, fontweight="bold")
fig.tight_layout(rect=[0, 0, 1, 0.97])
plt.show()

# Get outliers based on 0.1 0.99, |z| > 3

def num_less_than_1st_percentile(x):
    return x[x < x.quantile(0.01)].count()

def num_greater_than_99th_percentile(x):
    return x[x > x.quantile(0.99)].count()

def num_greater_than_3z(x):
    return x[zscore(x) > 3].count()

def num_less_than_neg3z(x):
    return x[zscore(x) < -3].count()

aggregated_outliers = numeric_df.agg([
    num_less_than_1st_percentile,
    num_greater_than_99th_percentile,
    num_greater_than_3z,
    num_less_than_neg3z
])

display(aggregated_outliers)

car_sales_df["estimate_diff"] = car_sales_df["sellingprice"] - car_sales_df["mmr"]
display(car_sales_df["estimate_diff"].describe())

count    472325.000000
mean       -146.546905
std        1741.175011
min      -87750.000000
25%        -800.000000
50%         -50.000000
75%         650.000000
max      207200.000000
Name: estimate_diff, dtype: float64

display(car_sales_df["estimate_diff"][abs(zscore(car_sales_df["estimate_diff"])) > 3].reset_index())
car_sales_df["estimate_diff"].hist(bins=50)

<Axes: >

# Clean
df = car_sales_df[["sellingprice", "estimate_diff"]]
x = df["sellingprice"].astype(float).values
y = df["estimate_diff"].astype(float).values
if df.empty:
    raise ValueError("No data to plot after dropping NaNs.")

# Fit
slope, intercept, r, p, se = linregress(x, y)

# Plot scatter + regression line
fig, ax = plt.subplots(figsize=(6, 4))
ax.scatter(x, y, alpha=0.6)
xs = np.linspace(x.min(), x.max(), 200)
ax.plot(xs, slope * xs + intercept, color="red", lw=1.5)

ax.set_title("Selling Price vs Estimate Diff", fontsize=10, fontweight="bold")
ax.set_xlabel("Selling Price")
ax.set_ylabel("Estimate Diff")
ax.grid(True, linestyle="--", alpha=0.3)

ax.text(
    0.02, 0.98,
    f"slope={slope:.3g}  R²={r**2:.3f}  p={p:.2e}  n={len(df)}",
    transform=ax.transAxes, va="top"
)

plt.tight_layout()
plt.show()

# ML Model for Car Sales Price Prediction
# Based on insights from previous analysis:
# 1. MMR is a strong predictor (positive linear relationship)
# 2. Age at sale is inversely correlated with price
# 3. Need to handle categorical features (make, model, body, transmission, state)
# 4. Outliers exist in odometer, sellingprice, age_at_sale

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, LabelEncoder
from sklearn.linear_model import LinearRegression, Ridge
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
import warnings
warnings.filterwarnings('ignore')

# Set random seed for reproducibility
random_state = 42
np.random.seed(random_state)

# Prepare a clean copy of the dataframe for ML
ml_df = car_sales_df.copy()

# Ensure age_at_sale is calculated (in case it wasn't in previous cells)
if "age_at_sale" not in ml_df.columns:
    ml_df["age_at_sale"] = ml_df["saledate"] - ml_df["year"]
    ml_df["age_at_sale_years"] = ml_df["age_at_sale"].dt.days / 365.25
else:
    ml_df["age_at_sale_years"] = ml_df["age_at_sale"].dt.days / 365.25


ml_df["year_numeric"] = ml_df["year"].dt.year

# Create additional features
# 1. Log of age (as we found log relationship in analysis)
ml_df["log_age_at_sale"] = np.log1p(ml_df["age_at_sale_years"])

# 2. Log of odometer (to handle right skewness)
ml_df["log_odometer"] = np.log1p(ml_df["odometer"])

# 3. Mileage per year (condition indicator)
ml_df["miles_per_year"] = ml_df["odometer"] / (ml_df["age_at_sale_years"] + 0.1)  # +0.1 to avoid division by zero

# Display basic info
print(f"Dataset shape: {ml_df.shape}")
print(f"\nMissing values:")
print(ml_df.isnull().sum()[ml_df.isnull().sum() > 0])

Dataset shape: (472325, 23)

Missing values:
Series([], dtype: int64)

# Select features for modeling
# Numerical features
numerical_features = [
    'mmr',  # Strong predictor from analysis
    'odometer',
    'condition',
    'age_at_sale_years',
    'log_age_at_sale',  # Log transformation based on analysis
    'log_odometer',
    'miles_per_year',
    'year_numeric'
]

# Categorical features (will be encoded)
categorical_features = [
    'make',
    'model',
    'body',
    'transmission',
    'state'
]

# Target variable
target = 'sellingprice'

# Check which features exist
available_numerical = [f for f in numerical_features if f in ml_df.columns]
available_categorical = [f for f in categorical_features if f in ml_df.columns]

print(f"Available numerical features: {available_numerical}")
print(f"Available categorical features: {available_categorical}")

# Create feature matrix and target
# For tree-based models, we can use label encoding for categoricals
# For linear models, we'd need one-hot encoding, but we'll use tree-based models primarily

# Prepare data - drop rows with missing target
ml_df_clean = ml_df.dropna(subset=[target]).copy()

print(f"\nClean dataset shape: {ml_df_clean.shape}")
print(f"Target variable statistics:")
print(ml_df_clean[target].describe())

Available numerical features: ['mmr', 'odometer', 'condition', 'age_at_sale_years', 'log_age_at_sale', 'log_odometer', 'miles_per_year', 'year_numeric']
Available categorical features: ['make', 'model', 'body', 'transmission', 'state']

Clean dataset shape: (472325, 23)
Target variable statistics:
count    472325.000000
mean      13690.512058
std        9613.033738
min           1.000000
25%        7200.000000
50%       12200.000000
75%       18200.000000
max      230000.000000
Name: sellingprice, dtype: float64

# Encode categorical variables using Label Encoding
# (Tree-based models can handle label-encoded categoricals well)
label_encoders = {}
ml_df_encoded = ml_df_clean.copy()

for col in available_categorical:
    le = LabelEncoder()
    ml_df_encoded[col + '_encoded'] = le.fit_transform(ml_df_encoded[col].astype(str))
    label_encoders[col] = le
    print(f"{col}: {len(le.classes_)} unique values")

# Create final feature list
feature_columns = available_numerical + [col + '_encoded' for col in available_categorical]

# Extract features and target
X = ml_df_encoded[feature_columns].copy()
y = ml_df_encoded[target].copy()

print(f"\nFeature matrix shape: {X.shape}")
print(f"Features used: {feature_columns}")

make: 53 unique values
model: 768 unique values
body: 85 unique values
transmission: 2 unique values
state: 34 unique values

Feature matrix shape: (472325, 13)
Features used: ['mmr', 'odometer', 'condition', 'age_at_sale_years', 'log_age_at_sale', 'log_odometer', 'miles_per_year', 'year_numeric', 'make_encoded', 'model_encoded', 'body_encoded', 'transmission_encoded', 'state_encoded']

# Handle any remaining missing values (shouldn't be many after cleaning)
print(f"Missing values in X: {X.isnull().sum().sum()}")
if X.isnull().sum().sum() > 0:
    # Fill with median for numerical, mode for categorical
    for col in X.columns:
        if X[col].isnull().sum() > 0:
            if '_encoded' in col:
                X[col].fillna(X[col].mode()[0], inplace=True)
            else:
                X[col].fillna(X[col].median(), inplace=True)

# Split data into train and test sets
test_size = 0.2
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=test_size, random_state=random_state, shuffle=True
)

print(f"Training set size: {X_train.shape[0]}")
print(f"Test set size: {X_test.shape[0]}")
print(f"\nTraining target statistics:")
print(y_train.describe())

Missing values in X: 0
Training set size: 377860
Test set size: 94465

Training target statistics:
count    377860.000000
mean      13696.451699
std        9632.775108
min           1.000000
25%        7200.000000
50%       12200.000000
75%       18200.000000
max      230000.000000
Name: sellingprice, dtype: float64

# Initialize models
# Using Linear Regression and Ridge Regression (regularized linear model)
models = {
    'Linear Regression': LinearRegression(),
    'Ridge Regression': Ridge(alpha=1.0, random_state=random_state)
}

# Train and evaluate models
results = {}

for name, model in models.items():
    print(f"\n{'='*60}")
    print(f"Training {name}...")

    # Train model
    model.fit(X_train, y_train)

    # Predictions
    y_train_pred = model.predict(X_train)
    y_test_pred = model.predict(X_test)

    # Calculate metrics
    train_r2 = r2_score(y_train, y_train_pred)
    test_r2 = r2_score(y_test, y_test_pred)
    train_mae = mean_absolute_error(y_train, y_train_pred)
    test_mae = mean_absolute_error(y_test, y_test_pred)
    train_rmse = np.sqrt(mean_squared_error(y_train, y_train_pred))
    test_rmse = np.sqrt(mean_squared_error(y_test, y_test_pred))

    results[name] = {
        'model': model,
        'train_r2': train_r2,
        'test_r2': test_r2,
        'train_mae': train_mae,
        'test_mae': test_mae,
        'train_rmse': train_rmse,
        'test_rmse': test_rmse,
        'y_test_pred': y_test_pred
    }

    print(f"Train R²: {train_r2:.4f} | Test R²: {test_r2:.4f}")
    print(f"Train MAE: ${train_mae:.2f} | Test MAE: ${test_mae:.2f}")
    print(f"Train RMSE: ${train_rmse:.2f} | Test RMSE: ${test_rmse:.2f}")

============================================================
Training Linear Regression...
Train R²: 0.9697 | Test R²: 0.9713
Train MAE: $1043.20 | Test MAE: $1037.53
Train RMSE: $1677.50 | Test RMSE: $1614.93

============================================================
Training Ridge Regression...
Train R²: 0.9697 | Test R²: 0.9713
Train MAE: $1043.20 | Test MAE: $1037.53
Train RMSE: $1677.50 | Test RMSE: $1614.92

# Create comparison DataFrame
comparison_df = pd.DataFrame({
    'Model': list(results.keys()),
    'Train R²': [results[m]['train_r2'] for m in results.keys()],
    'Test R²': [results[m]['test_r2'] for m in results.keys()],
    'Train MAE': [results[m]['train_mae'] for m in results.keys()],
    'Test MAE': [results[m]['test_mae'] for m in results.keys()],
    'Train RMSE': [results[m]['train_rmse'] for m in results.keys()],
    'Test RMSE': [results[m]['test_rmse'] for m in results.keys()]
})

# Sort by Test R² (best first)
comparison_df = comparison_df.sort_values('Test R²', ascending=False).reset_index(drop=True)

print("\n" + "="*80)
print("MODEL COMPARISON SUMMARY")
print("="*80)
display(comparison_df.round(4))

# Identify best model
best_model_name = comparison_df.iloc[0]['Model']
best_model = results[best_model_name]['model']
print(f"\n Best Model: {best_model_name}")
print(f"   Test R²: {comparison_df.iloc[0]['Test R²']:.4f}")
print(f"   Test MAE: ${comparison_df.iloc[0]['Test MAE']:.2f}")
print(f"   Test RMSE: ${comparison_df.iloc[0]['Test RMSE']:.2f}")

================================================================================
MODEL COMPARISON SUMMARY
================================================================================

 Best Model: Ridge Regression
   Test R²: 0.9713
   Test MAE: $1037.53
   Test RMSE: $1614.92

# Plot predictions vs actual for best model
best_pred = results[best_model_name]['y_test_pred']

fig, axes = plt.subplots(1, 2, figsize=(15, 5))

# Scatter plot: Predicted vs Actual
axes[0].scatter(y_test, best_pred, alpha=0.5, s=10)
axes[0].plot([y_test.min(), y_test.max()], [y_test.min(), y_test.max()], 'r--', lw=2, label='Perfect Prediction')
axes[0].set_xlabel('Actual Selling Price ($)', fontsize=12)
axes[0].set_ylabel('Predicted Selling Price ($)', fontsize=12)
axes[0].set_title(f'{best_model_name}: Predicted vs Actual', fontsize=14, fontweight='bold')
axes[0].legend()
axes[0].grid(True, alpha=0.3)

# Residual plot
residuals = y_test - best_pred
axes[1].scatter(best_pred, residuals, alpha=0.5, s=10)
axes[1].axhline(y=0, color='r', linestyle='--', lw=2)
axes[1].set_xlabel('Predicted Selling Price ($)', fontsize=12)
axes[1].set_ylabel('Residuals ($)', fontsize=12)
axes[1].set_title(f'{best_model_name}: Residual Plot', fontsize=14, fontweight='bold')
axes[1].grid(True, alpha=0.3)

plt.tight_layout()
plt.show()

# Print some statistics about residuals
print(f"\nResidual Statistics:")
print(f"Mean residual: ${residuals.mean():.2f}")
print(f"Std residual: ${residuals.std():.2f}")
print(f"95% of predictions within: ±${np.percentile(np.abs(residuals), 95):.2f}")

Residual Statistics:
Mean residual: $5.58
Std residual: $1614.92
95% of predictions within: ±$3010.13

	year	make	model	trim	body	transmission	vin	state	condition	odometer	color	interior	seller	mmr	sellingprice	saledate
0	2015	Kia	Sorento	LX	SUV	automatic	5xyktca69fg566472	ca	5.0	16639.0	white	black	kia motors america inc	20500.0	21500.0	Tue Dec 16 2014 12:30:00 GMT-0800 (PST)
1	2015	Kia	Sorento	LX	SUV	automatic	5xyktca69fg561319	ca	5.0	9393.0	white	beige	kia motors america inc	20800.0	21500.0	Tue Dec 16 2014 12:30:00 GMT-0800 (PST)
2	2014	BMW	3 Series	328i SULEV	Sedan	automatic	wba3c1c51ek116351	ca	45.0	1331.0	gray	black	financial services remarketing (lease)	31900.0	30000.0	Thu Jan 15 2015 04:30:00 GMT-0800 (PST)
3	2015	Volvo	S60	T5	Sedan	automatic	yv1612tb4f1310987	ca	41.0	14282.0	white	black	volvo na rep/world omni	27500.0	27750.0	Thu Jan 29 2015 04:30:00 GMT-0800 (PST)
4	2014	BMW	6 Series Gran Coupe	650i	Sedan	automatic	wba6b2c57ed129731	ca	43.0	2641.0	gray	black	financial services remarketing (lease)	66000.0	67000.0	Thu Dec 18 2014 12:30:00 GMT-0800 (PST)

	year	make	model	trim	body	transmission	vin	state	condition	odometer	color	interior	seller	mmr	sellingprice	saledate
0	2014-01-01 00:00:00+00:00	Kia	Sorento	LX	SUV	automatic	5xyktca69fg566472	ca	5.0	16639.0	white	black	kia motors america inc	20500.0	21500.0	2014-12-16 20:30:00+00:00
1	2014-01-01 00:00:00+00:00	Kia	Sorento	LX	SUV	automatic	5xyktca69fg561319	ca	5.0	9393.0	white	beige	kia motors america inc	20800.0	21500.0	2014-12-16 20:30:00+00:00
2	2013-01-01 00:00:00+00:00	BMW	3 Series	328i SULEV	Sedan	automatic	wba3c1c51ek116351	ca	45.0	1331.0	gray	black	financial services remarketing (lease)	31900.0	30000.0	2015-01-15 12:30:00+00:00
3	2014-01-01 00:00:00+00:00	Volvo	S60	T5	Sedan	automatic	yv1612tb4f1310987	ca	41.0	14282.0	white	black	volvo na rep/world omni	27500.0	27750.0	2015-01-29 12:30:00+00:00
4	2013-01-01 00:00:00+00:00	BMW	6 Series Gran Coupe	650i	Sedan	automatic	wba6b2c57ed129731	ca	43.0	2641.0	gray	black	financial services remarketing (lease)	66000.0	67000.0	2014-12-18 20:30:00+00:00

	make	slope	intercept	r_squared	p_value	std_err	samples
0	Acura	-14367.574927	126111.689156	0.814467	0.000000e+00	102.212081	4503
20	Audi	-19077.368981	165925.241154	0.569782	0.000000e+00	239.269987	4802
21	BMW	-19749.804437	172113.236597	0.700462	0.000000e+00	97.609277	17509
39	Bentley	-71353.692615	640267.166406	0.902505	7.061489e-54	2310.810318	105
23	Buick	-10436.566516	91630.728406	0.720159	0.000000e+00	97.328952	4470
24	Cadillac	-17926.827616	155979.125246	0.678586	0.000000e+00	155.279999	6315
25	Chevrolet	-8347.427233	74653.150820	0.356957	0.000000e+00	48.147473	54150
18	Chrysler	-8315.417309	72555.290940	0.671989	0.000000e+00	47.229463	15133
28	Dodge	-7042.701755	63558.198314	0.434857	0.000000e+00	48.699916	27181
37	FIAT	-5489.014168	48271.330264	0.400099	2.792176e-82	249.966045	725
30	Ford	-8677.152345	78543.244428	0.346703	0.000000e+00	41.848657	81013
31	GMC	-12879.890144	115506.376188	0.628810	0.000000e+00	103.086751	9217
47	HUMMER	-4097.152084	49020.044155	0.009001	8.515933e-03	1553.294524	768
17	Honda	-9569.238356	84674.646012	0.683540	0.000000e+00	41.363074	24781
7	Hyundai	-7181.888171	63545.545978	0.470704	0.000000e+00	55.756201	18659
1	Infiniti	-13935.160695	123645.291972	0.617767	0.000000e+00	92.610576	14011
41	Isuzu	-4631.096168	41339.797039	0.454137	4.761581e-25	381.632354	179
2	Jaguar	-29224.390020	250488.327279	0.826279	0.000000e+00	378.256206	1257
3	Jeep	-10273.131802	92425.789893	0.510926	0.000000e+00	89.073015	12735
4	Kia	-7344.054009	64018.948016	0.637345	0.000000e+00	44.037584	15827
6	Land Rover	-36539.858078	310061.091441	0.728927	0.000000e+00	613.310900	1322
8	Lexus	-16119.859951	143567.426007	0.738760	0.000000e+00	93.965133	10409
15	Lincoln	-14088.830956	122971.778703	0.792892	0.000000e+00	102.614097	4926
9	MINI	-9815.922787	86054.669203	0.774553	0.000000e+00	97.602092	2946
40	Maserati	-32502.748799	294607.519084	0.667091	2.609532e-27	2219.719205	109
10	Mazda	-7718.192359	67895.757754	0.659698	0.000000e+00	66.599543	6930
11	Mercedes-Benz	-18628.337216	163993.457608	0.546134	0.000000e+00	143.493247	14008
12	Mercury	-7773.926405	68129.739890	0.621654	0.000000e+00	153.108510	1571
13	Mitsubishi	-6556.421913	57960.909732	0.579227	0.000000e+00	92.381552	3661
14	Nissan	-6834.869326	62470.826918	0.375004	0.000000e+00	42.045754	44043
43	Oldsmobile	-1390.224169	12952.969599	0.088056	6.936989e-08	251.677897	318
27	Pontiac	-6480.339369	56997.453668	0.326874	0.000000e+00	146.743598	4018
33	Porsche	-35753.658373	313592.698484	0.688749	5.280718e-295	707.219499	1157
34	Ram	-12098.213431	110741.388667	0.220257	1.149539e-220	358.041813	4044
36	Saab	-8840.661982	76610.424274	0.605477	1.225837e-85	350.729255	416
16	Saturn	-7280.853106	63461.497051	0.513983	0.000000e+00	139.712889	2570
26	Scion	-8193.130217	71416.061663	0.751657	0.000000e+00	119.890141	1545
29	Subaru	-9191.259827	82660.039417	0.709713	0.000000e+00	88.951962	4369
35	Suzuki	-5445.477932	47248.270180	0.563907	6.701890e-178	153.362616	977
22	Toyota	-7415.130205	68160.186477	0.423977	0.000000e+00	45.995233	35313
19	Volkswagen	-7456.066833	65829.813164	0.651636	0.000000e+00	52.511543	10780
5	Volvo	-11036.399336	96537.451790	0.880310	0.000000e+00	73.326795	3082
38	smart	-3879.416386	34910.424049	0.560402	7.960942e-63	186.065995	343

	odometer	sellingprice	age_at_sale	year
num_less_than_1st_percentile	4724	3825	4552	4145
num_greater_than_99th_percentile	4724	4656	4703	0
num_greater_than_3z	4550	5841	4473	0
num_less_than_neg3z	0	0	0	4145

	index	estimate_diff
0	9	-8800.0
1	20	-17100.0
2	40	-5400.0
3	56	-6800.0
4	59	6250.0
...	...	...
6788	558777	-13050.0
6789	558779	-15000.0
6790	558791	-7100.0
6791	558807	5500.0
6792	558826	-12500.0

Used Vehicle Sales Price Prediction

Contributions

Introduction

Data Curation

Data preprocessing

Cleaning

Exploratory Data Analysis

Conclusion #1 - The Relationship Between Age-at-sale and Selling Price

Relationship within manufacturers

Conclusion #2 - Checking for Outliers in Numeric-like Features

Format the data

Create the plots

Conclusion #3 - Investigating MMR vs Selling Price

Primary Analysis

Feature Engineering and Data Preparation¶

Model Training and Evaluation¶

Visualization

Insights and Conclusion

	year	make	model	trim	body	transmission	vin	state	condition	odometer	color	interior	seller	mmr	sellingprice	saledate
14	2014	Chevrolet	Cruze	2LT	Sedan	automatic	1g1pe5sbxe7120097	ca	NaN	15686.0	blue	black	avis rac/san leandro	13900.0	10600.0	Tue Dec 16 2014 12:00:00 GMT-0800 (PST)
16	2015	Hyundai	Sonata	SE	Sedan	automatic	5npe24af4fh001562	ca	NaN	8311.0	red	—	avis tra	15200.0	4200.0	Tue Dec 16 2014 13:00:00 GMT-0800 (PST)
22	2014	Chevrolet	Camaro	LT	Convertible	automatic	2g1fb3d31e9134662	ca	NaN	33450.0	black	black	avis rac/san leandro	20100.0	14700.0	Tue Dec 16 2014 12:00:00 GMT-0800 (PST)
25	2015	Hyundai	Sonata	SE	Sedan	automatic	5npe24af4fh038482	ca	NaN	9281.0	silver	gray	enterprise vehicle exchange / tra / rental / t...	15150.0	8500.0	Tue Dec 16 2014 13:00:00 GMT-0800 (PST)
28	2014	BMW	X5	sDrive35i	SUV	automatic	5uxkr2c52e0h33130	ca	NaN	11278.0	gray	black	avis rac/san leandro	50400.0	34000.0	Tue Dec 16 2014 13:00:00 GMT-0800 (PST)

	condition	odometer	mmr	sellingprice
count	472325.000000	472325.000000	472325.000000	472325.000000
mean	30.774177	66701.732040	13837.058964	13690.512058
std	13.286866	51939.586894	9532.229273	9613.033738
min	1.000000	1.000000	25.000000	1.000000
25%	24.000000	28137.000000	7425.000000	7200.000000
50%	35.000000	51085.000000	12300.000000	12200.000000
75%	41.000000	96590.000000	18300.000000	18200.000000
max	49.000000	999999.000000	182000.000000	230000.000000

	Model	Train R²	Test R²	Train MAE	Test MAE	Train RMSE	Test RMSE
0	Ridge Regression	0.9697	0.9713	1043.2025	1037.5325	1677.5018	1614.9248
1	Linear Regression	0.9697	0.9713	1043.2029	1037.5328	1677.5018	1614.9250